RESArnau opened a new issue, #36551:
URL: https://github.com/apache/airflow/issues/36551

   ### Apache Airflow version
   
   Other Airflow 2 version (please specify below)
   
   ### If "Other Airflow 2 version" selected, which one?
   
   2.5.3
   
   ### What happened?
   
   In our Google Cloud Composer environment, we have been experiencing a series 
of unexpected task failures in result of a task mark as zombie. These failures 
do not appear to correlate with system resource constraints, as both CPU and 
memory usage are below their respective limits according to our monitoring 
dashboards. The task is just waiting for a message in a pub/sub queue. It only 
happens in one of our two enviroments. 
   
   ### What you think should happen instead?
   
   Tasks should keep beeing executed without being marked as zombie when system 
resources are underutilized, and no other apparent issues are present that 
would cause such failures.
   
   ### How to reproduce
   
   The issue can be sporadically observed in our Cloud Composer environment 
when executing tasks. It is not consistently reproducible but occurs frequently 
enough to impact our workflows.
   
   ### Operating System
   
   composer-2.5.1
   
   ### Versions of Apache Airflow Providers
   
   _No response_
   
   ### Deployment
   
   Google Cloud Composer
   
   ### Deployment details
   
   composer-2.5.1
   
   Scheduler:
   2 schedulers with 1.5 vCPU, 3 GB memory, 1 GB storage each
   
   Triggerer:
   1 triggerer with 1 vCPU, 1 GB memory, 1 GB storage
   
   Web server:
   1 vCPU, 3 GB memory, 1 GB storage
   
   Worker:
   Autoscaling between 1 and 6 workers, with 1.5 vCPU, 3 GB memory, 1 GB 
storage each
   
   Environment size:
   Small
   
   ### Anything else?
   
   _No response_
   
   ### Are you willing to submit PR?
   
   - [ ] Yes I am willing to submit a PR!
   
   ### Code of Conduct
   
   - [X] I agree to follow this project's [Code of 
Conduct](https://github.com/apache/airflow/blob/main/CODE_OF_CONDUCT.md)
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to